Support Vector Machine In Traffic Sign Recognition

With the fast development of society and economy, the jamming and huddling of traffic become more and more serious, and become one of the bottle-neck of our modern city, traffic accidents are paid much attention by the governments of the world. In this kind of conditions, Intelligent Traffic System (ITS) is come into being. It involves the technologies of pattern recognition, digital image processing, artificial intelligence, electronic technology, communication technology, system engineering and so no. At present, many countries are busy in the researching and erecting of their own ITS.The traffic sign recognition (TSR) the important component of intelligent vehicle, it collecting and recognition the road sign information in the vehicle driving process, then gave alert or warnings to driver, or control the operation of the vehicle directly, to keep the transportation smoothing and avoiding traffic accident: The automatic segmentation of road sign and recognition is the important support software of the intelligent transportation system, there are important theory and practical values.The keystone and difficulty lies in the segment of traffic sign, feature abstracting and the design of classifier.In the research of traffic segment, through the analysis of the traffic sign characters, a segment way based on RGB model is proposed. In RGB model, the red ,green and blue which are belong to the same pixel are easy to be influenced by illumination, but the values of theirs difference are keep in a fixed bound and are not influenced by illumination. The way is free from lighting and the change of model is not necessary, so the speed of the algorithm is improved. Based on this theory, the paper selects some traffic sign and segments them with the segment way that we propose. In the experiment, we find the segment effect is good.Because our final propose aims at the traffic sign recognition, it is very important to select some features that can represent the characters of traffic sign. When Object's size is zoom out or zoom in, or its position is changed, or it is circumrotated, but its moment invariants do not change. So the moment invariants of object are usually used in target recognition. In this paper we are research on the theory of moment invariants firstly, and then find ten new moment invariants by deduction. New moment invariants also have the property that moment invariants don't change, in despite of object's size, position is change and object iscircumrotated. Because we throw off u₀₀ in moment invariants formula, so the newmoment invariants have no business with object's area and structure and can represent the characters of the object whose structure is close and disclose. Finally the paper selects the moment invariants as the characters of traffic sign, and distills the moment invariants of some traffic sign.Because the features of SVM and the potential in traffic sign recognition, we bring forward a recognition modes based on decision tree SVM after plenty of researching. The paper does a lot of experiment on SVM train and recognition, gains an excellent SVM model. We use the model in our TRS.